Methodology for Choosing Optimal Parameters for Real-time MPC in Urban Drainage Systems

Abstract

The main goal of Urban Drainage Systems (UDS) is to properly handle rainwater, surface runoff, and sewage, minimising overflow and other health hazards. Combined Sewer Overflow (CSO) events are still a major challenge in UDS management, and many software techniques are used to reduce overflow events, for example Model Predictive Control (MPC). By applying an optimisation routine using a model of the UDS, MPC can reduce the impact or even avoid CSO events. However, for this method to be effective, it requires the selection of several parameters, which is typically done through trial-and-error. This paper proposes a methodology to identify the MPC’s parameters by using a Sensitivity Analysis (SA) combined with a Genetic Algorithm (GA). The methodology consists of identifying regions of effectiveness of the parameters for desired outputs using the SA and applying the knowledge into a GA which finds the best parameters. The results showed that SA reduced the computational cost for the GA while providing an excellent combination of parameters. The novel methodology is applied to a simple network with 3 CSO tanks, reducing the system’s total volume overflow over a 1-year period by 14% compared to when MPC is not implemented.